Brooks's Law Refuses to Die: Managing AI Agents Is the New Career Skill

Fifty-one years after Frederick Brooks wrote The Mythical Man-Month, one of software’s oldest laws is doing fresh work in the AI era — and this week it became the most-read career story on Hacker News. Jan van den Berg’s essay “The Mythical Man-Month in the age of AI”, published 27 October, climbed to 369 points with the argument that Brooks’s Law hasn’t been repealed by agents. It has mutated.

Brooks’s original formulation, drawn from managing IBM’s OS/360 project: adding manpower to a late software project makes it later. More people means more communication overhead than the extra hands repay. The obvious 2026 assumption is that the law is dead — an AI agent needs no onboarding, no stand-ups, no coffee. Van den Berg’s essay, read widely on Hacker News overnight, says otherwise: the overhead never vanished, it moved.

🔍 THE BOTTOM LINE: The career skill AI has made scarce is not prompt-writing — it is the judgement to define what “done” means, review machine output, catch plausible hallucinations, and stitch fragments into a working whole. Two essays topping Hacker News this week converge on the same profile: the person managing the agents is the job. For workers deciding what to be good at, that is the signal in the data.

The law, updated

Nine agents still can’t deliver a baby in one month — sequential tasks stay sequential no matter how fast the workers are. What changes with agents is where the coordination cost lands. Van den Berg’s 2026 update, as the essay puts it:

“Adding agents to a project with unclear requirements gets you the wrong answer faster.”

That reframing matters for anyone planning a career around AI tools. The essay’s core observation: instead of managing people, you are now managing agents — someone still has to define “done”, review the code, catch hallucinations, and stitch the pieces together. “That someone is you.”

It echoes a pattern Singularity.Kiwi has tracked across 2026’s hiring data: job postings that reference AI skills up 93 percent in New Zealand year-on-year, even as total advertised vacancies stayed flat — employers want AI capability inside existing roles, not a separate priesthood running the machines. The essay lands on the same ground from the other direction: the scarce skill is orchestration plus review, and it sits inside every technical role.

Not a new claim — but newly payrolled

None of this is conceptually novel. What is new in October 2026 is how explicitly employers are hiring for it. Three data points from this month alone:

Microsoft opened a voluntary exit offer to roughly 60,000 employees on 26 October, its biggest headcount tool ever — while reporting says AI infrastructure spending is what the savings must fund. The people staying are being reorganised around smaller teams with more machine assistance.
Netherlands plans to take 25,000 civil servants off the payroll by 2030, with AI handling part of the workload — and, critically, new supervisory roles to check what it does.
Indeed’s own reporting this quarter showed AI job postings up but junior hires flat, the split the “agents need managers” thesis predicts: fewer entry seats, more senior seats holding the clipboard.

The essay’s practical warning is to teams, not individuals: nine autonomous agents turned loose on one codebase without clear orchestration “will happily rewrite each other’s work in a high-speed loop.” The fix is a human with authority over requirements and review — which is, structurally, a management job. Management jobs have historically been the ones companies actually pay for.

The positive read for workers

The site has published plenty on what AI removes — the vanishing apprenticeship first rung among them. What the Brooks’s Law revival adds is a creation story with numbers behind it: the coordination layer that used to be staffed by project managers, tech leads and QA reviewers is being re-staffed, not eliminated, and the job descriptions are being written around agents.

For New Zealand readers, the angle is less about local big tech (there isn’t one at Amazon scale) and more about what small teams can do. NZ software shops and agencies compete on seniority, not headcount — a working model where one experienced person directs a fleet of agents is, bluntly, the business model many already run, just with better tooling this year. The essay’s advice to cultivate the ability to define, review and integrate — rather than to type faster — costs nothing, and maps to the skills NZ job ads already ask for.

The debate the essay reopened — whether agents genuinely make teams faster at all — remains genuinely unsettled; a recent analysis of token overhead costs found agent-first development can cost more than the humans it replaces. Brooks’s Law mutating doesn’t mean it’s gone. It means the person who spots the mutation earliest gets paid for it.

❓ FAQ

What is Brooks’s Law? That adding manpower to a late software project makes it later — formulated by Frederick P. Brooks Jr. in his 1975 book The Mythical Man-Month (Wikipedia), based on his experience managing IBM’s OS/360 project.

Why is Brooks’s Law relevant to AI agents? Because coordination overhead doesn’t disappear when the workers are machines — the essay argues it shifts to defining requirements, reviewing output, catching hallucinations and integrating fragments, all still human work.

What is the “2026 update” to Brooks’s Law? The essay’s phrasing: adding agents to a project with unclear requirements gets you the wrong answer faster.

Which jobs does this create? Review, orchestration and specification roles — “AI agent manager” and “agent orchestrator” titles are appearing on job boards in 2026, though most demand is folded into existing software and analysis roles rather than new titles.